{"id":"W7117365168","doi":"10.1016/j.imed.2025.12.006","title":"Evaluation of artificial intelligence-based tool Covidence in literature screening for guideline updates: A prospective study","year":2025,"lang":"en","type":"article","venue":"Intelligent Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hamilton Health Sciences; Population Health Research Institute; McMaster University; University of British Columbia; Impact","funders":"Hamilton Health Sciences Foundation","keywords":"Guideline; Randomized controlled trial; Workload; Breast cancer; MEDLINE; Lung cancer screening; Stage (stratigraphy); Clinical Practice; Evidence-based medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01462834,0.0002551309,0.0006814871,0.0007596163,0.00005691348,0.00001977129,0.0001883807,0.0001266433,0.0003661791],"category_scores_gemma":[0.0480476,0.0002065702,0.000121056,0.001520839,0.0001456978,0.0001631544,0.00005255303,0.0003703864,0.000005145141],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003865865,"about_ca_system_score_gemma":0.0008500788,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000325865,"about_ca_topic_score_gemma":0.000525796,"domain_scores_codex":[0.9942036,0.0003881651,0.003170482,0.0005880389,0.001371183,0.0002785765],"domain_scores_gemma":[0.9920776,0.002318703,0.0005305241,0.0005112046,0.004484197,0.00007780069],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006603979,0.002995394,0.0755818,0.0003902559,0.0004449692,0.00002877961,0.003604528,0.007335399,0.002390442,0.002087172,0.003456555,0.8950807],"study_design_scores_gemma":[0.01325676,0.0102101,0.03145885,0.008452484,0.005541197,0.00001511665,0.03617596,0.815287,0.03308838,0.04331239,0.002598407,0.0006033724],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1940765,0.001172085,0.7781123,0.01621666,0.0006610186,0.0094565,0.0000256231,0.00005627094,0.0002230459],"genre_scores_gemma":[0.984064,0.00003940028,0.01268462,0.002097831,0.0003253583,0.0004758534,0.0002384943,0.0000210404,0.00005341551],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8944774,"threshold_uncertainty_score":0.9599711,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3663738052635487,"score_gpt":0.5644291767119313,"score_spread":0.1980553714483825,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}